Landslide Hazard Prediction Map Based on Logistic Regression Model for Applying in the Whole Country of South Korea

Journal of the Korean Society of Hazard Mitigation 2014;14(6):117-123.
Published online December 31, 2014.

로지스틱 회귀모형을 이용한 전국 산사태 발생위험 예측지도 개발

우충식, 권현정, 이창우, 김경하

Landslide Hazard Prediction Map Based on Logistic Regression Model for Applying in the Whole Country of South Korea

Abstract

This study was carried out to develop landslide hazard prediction map for the whole country of South Korea using the data 1,912landslides that were occurred during the period of 2005 to 2011. These data set were extracted in the aerial photographs. Logisticregression model was developed using the data mentioned above. We built spatial database on landslide of the whole country using aerial photographs because the model based on probability such as logistic regression analysis needs a lot of data concerned on landslide.Nine factors closely related to landslide were selected through logistic regression analysis and the classification accuracy ofmodel were 75.3%. Applied factors are as follows; slope, drainage length, aspect, curvature, TWI, mother rock, soil depth, forest stand, tree diameter. Using the developed model and GIS analysis, The new landslide hazard prediction map of South Korea was made. For the validation of this map, we compared dangerous areas by model to 276 landslide areas that were occurred from 2012 to 2013. As a result, the hit rate was shown in 76% approximately.

요지

본 연구에서는 2005년부터 2011년까지 전국 산사태 발생지역을 촬영한 항공영상에서 추출한 산사태 발생지 1,912개소와 확률기반의 로지스틱 회귀모형을 이용하여 전국단위 산사태 발생위험 예측지도를 개발하였다. 로지스틱 회귀모형과 같은 통계기반의 모형은 다량의 관련 데이터가 필요하기 때문에 최근에 발생한 산사태 발생지의 항공영상을 이용하여 전국의 산사태 공간자료를 DB화하였다. 산사태 위험도 예측모형은 산사태 발생에 영향을 미칠 수 있는 지형, 수문, 식생, 토양, 지질 등 9개 인자를 반영하였으며 모형의 분류정확도는 75.3%로 나타났다. 개발된 모형과 GIS를 이용하여 전국의 산사태 발생위험 예측지도를 제작하였으며 2012년, 2013년 산사태 발생지 276개소를 대상으로 적중률을 분석한 결과 약 76%로 나타났다.